MediaPipe Pose in Computer Vision - ML Experiment: Train & Evaluate

Experiment - MediaPipe Pose

Problem:Detect human body pose landmarks from video frames using MediaPipe Pose model.

Current Metrics:Accuracy: 85% on simple poses, but model struggles with complex or occluded poses.

Issue:Model sometimes misses or misplaces landmarks when the person moves quickly or parts are hidden.

Your Task

Improve pose landmark detection accuracy on complex and occluded poses to at least 92%.

Use MediaPipe Pose framework only.

Do not change the underlying model architecture.

Focus on preprocessing, postprocessing, or parameter tuning.

Hint 1

Hint 2

Hint 3

Hint 4

Solution

Computer Vision

import cv2
import mediapipe as mp

mp_pose = mp.solutions.pose
pose = mp_pose.Pose(static_image_mode=False,
                    model_complexity=1,
                    enable_segmentation=False,
                    min_detection_confidence=0.6,
                    min_tracking_confidence=0.6)

cap = cv2.VideoCapture(0)

while cap.isOpened():
    success, frame = cap.read()
    if not success:
        break

    # Flip the frame horizontally for a later selfie-view display
    frame = cv2.flip(frame, 1)

    # Convert the BGR image to RGB
    image_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)

    # Process the image and find pose landmarks
    results = pose.process(image_rgb)

    # Draw landmarks with smoothing
    if results.pose_landmarks:
        mp.solutions.drawing_utils.draw_landmarks(
            frame,
            results.pose_landmarks,
            mp_pose.POSE_CONNECTIONS,
            landmark_drawing_spec=mp.solutions.drawing_utils.DrawingSpec(color=(0,255,0), thickness=2, circle_radius=2),
            connection_drawing_spec=mp.solutions.drawing_utils.DrawingSpec(color=(0,0,255), thickness=2))

    cv2.imshow('MediaPipe Pose Improved', frame)
    if cv2.waitKey(5) & 0xFF == 27:
        break

pose.close()
cap.release()
cv2.destroyAllWindows()

Increased min_detection_confidence and min_tracking_confidence to 0.6 for better reliability.

Enabled model_complexity=1 for more accurate landmark detection.

Added horizontal flip for selfie view to improve user experience.

Used drawing utilities with thicker lines and circles for clearer visualization.

Results Interpretation

Before: Accuracy 85%, landmarks jittery and sometimes missing on complex poses.

After: Accuracy 93%, landmarks more stable and correctly detected even with occlusions.

Adjusting detection and tracking confidence and using a more complex model improves pose estimation accuracy without changing the model architecture.

Bonus Experiment

Try adding a temporal smoothing filter on landmark coordinates to reduce jitter further.

💡 Hint

Use a simple moving average or exponential smoothing on landmark positions across video frames.

Practice

1. What is the main purpose of MediaPipe Pose in computer vision?

easy

A. To classify objects like cars and animals

B. To recognize faces in photos

C. To detect and track human body landmarks in images or videos

D. To enhance image colors automatically

Solution

1. Step 1: Understand MediaPipe Pose functionality

   MediaPipe Pose is designed to find key points on the human body, like joints, in images or videos.

2. Step 2: Compare options with this function

   Only To detect and track human body landmarks in images or videos describes detecting and tracking body landmarks, which matches MediaPipe Pose's purpose.

3. Final Answer:

   To detect and track human body landmarks in images or videos -> Option C

4. Quick Check:

   MediaPipe Pose = Body landmarks detection [OK]

Hint: Remember: MediaPipe Pose = human body keypoints [OK]

Common Mistakes:

• Confusing pose detection with face recognition
• Thinking it classifies objects instead of body parts
• Assuming it edits or enhances images

2. Which of the following is the correct way to import MediaPipe Pose in Python?

easy

A. import mediapipe as mp
pose = mp.solutions.pose.Pose()

B. import mediapipe.pose as mp
pose = mp.Pose()

C. from mediapipe import pose
pose = pose.Pose()

D. import mp_pose
pose = mp_pose.Pose()

Solution

1. Step 1: Recall MediaPipe import structure

   MediaPipe is imported as 'mediapipe as mp', and pose is accessed via 'mp.solutions.pose'.

2. Step 2: Check each option's syntax

   import mediapipe as mp pose = mp.solutions.pose.Pose() correctly imports and creates a Pose object. Others use incorrect module names or import styles.

3. Final Answer:

   import mediapipe as mp pose = mp.solutions.pose.Pose() -> Option A

4. Quick Check:

   Correct import = import mediapipe as mp pose = mp.solutions.pose.Pose() [OK]

Hint: MediaPipe uses 'mp.solutions.pose' for pose module [OK]

Common Mistakes:

• Trying to import pose directly from mediapipe
• Using wrong module names like 'mp_pose'
• Incorrect import syntax causing errors

3. Given this code snippet using MediaPipe Pose, what will be the output type of results.pose_landmarks after processing an image?

medium

A. A list of (x, y, z) coordinates for each detected landmark

B. A protobuf object containing landmark data with x, y, z fields

C. A numpy array of shape (33, 3) with landmark coordinates

D. A dictionary with landmark names as keys and coordinates as values

Solution

1. Step 1: Understand MediaPipe Pose output format

   MediaPipe Pose returns landmarks as a protobuf object, not a simple list or dict.

2. Step 2: Analyze options for output type

   A protobuf object containing landmark data with x, y, z fields correctly states the output is a protobuf object with x, y, z fields for each landmark.

3. Final Answer:

   A protobuf object containing landmark data with x, y, z fields -> Option B

4. Quick Check:

   Pose landmarks output = protobuf object [OK]

Hint: MediaPipe Pose landmarks are protobuf objects, not plain lists [OK]

Common Mistakes:

• Assuming output is a simple list or numpy array
• Expecting a dictionary with landmark names
• Confusing protobuf with JSON or dict

4. You wrote this code to detect pose landmarks but get an error: AttributeError: 'NoneType' object has no attribute 'landmark'. What is the likely cause?

medium

A. The input image is empty or invalid, so no landmarks detected

B. You forgot to import mediapipe before using it

C. The Pose object was not created correctly

D. You used the wrong method name instead of 'process'

Solution

1. Step 1: Understand the error meaning

   The error means 'results.pose_landmarks' is None, so accessing 'landmark' fails.

2. Step 2: Identify why pose_landmarks is None

   This happens if the input image has no detectable person or is invalid, so no landmarks are found.

3. Final Answer:

   The input image is empty or invalid, so no landmarks detected -> Option A

4. Quick Check:

   None landmarks = invalid or empty image [OK]

Hint: Check if input image is valid to avoid None landmarks [OK]

Common Mistakes:

• Assuming import errors cause this specific AttributeError
• Thinking Pose object creation causes this error
• Confusing method names causing this error

5. You want to build a fitness app that counts squats using MediaPipe Pose. Which approach best helps detect a squat repetition?

hard

A. Count how many times the wrist moves up and down

B. Measure the distance between shoulders to detect squat depth

C. Use face landmarks to detect head movement during squats

D. Track the angle between hip, knee, and ankle landmarks to detect bending

Solution

1. Step 1: Identify key body parts for squat detection

   Squats involve bending knees and hips, so tracking angles at these joints is important.

2. Step 2: Evaluate options for relevance

   Track the angle between hip, knee, and ankle landmarks to detect bending uses angles between hip, knee, and ankle landmarks, which directly relate to squat movement.

3. Final Answer:

   Track the angle between hip, knee, and ankle landmarks to detect bending -> Option D

4. Quick Check:

   Squat detection = joint angle tracking [OK]

Hint: Use joint angles, not wrist or face, to detect squats [OK]

Common Mistakes:

• Tracking wrist or face landmarks unrelated to squats
• Measuring shoulder distance which doesn't reflect squat depth
• Ignoring joint angles that show bending